Introduction

Context

  1. Data Preparation
  2. Part 1: Display of Gestures
  3. Part 2: Piecewise Aggregate Approximation
  4. Part 2: Adaptive Piecewise Representation with Linear Regressor
  5. Part 2: Gesture Recognation Perspective

Data Preparation

Gesture Filter

Part 1: Display

Graph Display Function

Gesture 1

Plotly Plot

In order to analyze which value (velocity or position) was analyzed by looking with an interactive plot.

Gesture 2

Gesture 3

Gesture 4

Gesture 5

Gesture 6

Gesture 7

Gesture 8

Part 2: 2 Representation Method

2 Different will be tried to represent existing dataset.

Piecewise Aggregate Approximation Function

There is 315 different time point in each axis, so it is hard to observe and analyze it by using Symbolic Aggregate approXimation(SAX) without Piecewise Aggregate Approximation(PAA). In this manner, SAX will be used with PAA in order to understand the symbols more easily. In addition, each axis will be constructed seperately.

Gesture 1 SAX with PAA

Hard Voting

Visual Control Gesture 1

One example implementation for new lines will be demonstrated at below. Remaning will be added to visualitation function in order to reduce repated codes for other Gestures.

Example Dataframe Construction

Fill of Na Values

After showing each step, other pictures will be obtained by using visualization function.

Visualization by function

All Gestures' axes will be controlled for 1 item in order to observe any problem for a specific gesture(like rapid changes). The main control of this is to understand the performance of the PAA in one dimension. The general performance of the function will be determined after controlling the second alternative on one axis visualization.

X axis

Y axis

Z axis

Gesture 2 SAX with PAA

Hard Voting

X axis

Y axis

Z axis

Gesture 3 SAX with PAA

Hard Voting

X axis

Y axis

Z axis

Gesture 4 SAX with PAA

Hard Voting

X axis

Y axis

Z axis

Gesture 5 SAX with PAA

Hard Voting

X axis

Y axis

Z axis

Gesture 6 SAX with PAA

Hard Voting

X axis

Y axis

Z axis

Gesture 7 SAX with PAA

Hard Voting

X axis

Y axis

Z axis

Gesture 8 SAX with PAA

Hard Voting

X axis

Y axis

Z axis

General Comment for Axis Representation

If there is an huge increase or decrease it is hard to represent this period by a single line. This problem can be handled by increasing the number of slot bu it can cause an overfitting problem. In addition, it is hard to understand the gesture by seperately looking the axis information. Final Gesture recognition will be handled at the end of the notebook by using moderate hard majority voting technique. In addition, numerical values didn't calculated because only visual inspection was demanded in homework description.

Adaptive Piecewise Representation with Linear Regressor

Gesture 1 with Regression Tree

Individual Decision Models

PCA for Gesture 1

Display of Axis

Gesture 2 with Regression Tree

Individual Decision Models

PCA for Gesture 2

Display of Axis

Gesture 3 with Regression Tree

Individual Decision Models

PCA for Gesture 3

Display of Axis

Gesture 4 with Regression Tree

Individual Decision Models

PCA for Gesture 4

Display of Axis

Gesture 5 with Regression Tree

Individual Decision Models

PCA for Gesture 5

Display of Axis

Gesture 6 with Regression Tree

Individual Decision Models

PCA for Gesture 6

Display of Axis

Gesture 7 with Regression Tree

Individual Decision Models

PCA for Gesture 7

Display of Axis

Gesture 8 with Regression Tree

Individual Decision Models

PCA for Gesture 8

Display of Axis

Performance of Regressor

Regression Tree to Find Piecewise Models on PCA

Neden bunu yaptım

Gesture Recognation Perspective

String to Value

DataFrame for String

Reduced Gesture Recognition

DataFrame for Regressors

Value Normalization

DataFrame for String

Reduced Gesture Recognition